MIT Robotics Pioneer Rodney Brooks Thinks People Are Vastly Overestimating Generative AI

Cover Image

When it comes to robotics and artificial intelligence, Rodney Brooks is a name to be reckoned with. As the Panasonic Professor of Robotics Emeritus at MIT, he has co-founded three key companies, including Rethink Robotics, iRobot, and his current venture, Robust.ai. Brooks also ran the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) for a decade starting in 1997.

Brooks is known for making predictions about the future of AI and keeps a scorecard on his blog to track his accuracy. He believes that people are vastly overestimating the capabilities of generative AI. "I'm not saying LLMs are not important, but we have to be careful with how we evaluate them," he said.

The issue with generative AI, according to Brooks, is that it can perform certain tasks well, but it can't do everything a human can. Humans tend to overestimate its capabilities because they generalize its performance to similar tasks. "When a human sees an AI system perform a task, they immediately generalize it to things that are similar and make an estimate of the competence of the AI system; not just the performance on that, but the competence around that," Brooks explained.

Brooks uses his company, Robust.ai, as an example. Someone suggested using a large language model (LLM) to tell his warehouse robots where to go, but Brooks believes this is not a reasonable use case for generative AI. Instead, it's simpler to connect the robots to a stream of data coming from the warehouse management software.

"When you have 10,000 orders that just came in that you have to ship in two hours, you have to optimize for that. Language is not going to help; it's just going to slow things down," he said. "We have massive data processing and massive AI optimization techniques and planning. And that's how we get the orders completed fast."

Brooks has also learned that you can't try to do too much with robots and AI. You should solve a solvable problem where robots can be integrated easily. "We need to automate in places where things have already been cleaned up. So the example of my company is we're doing pretty well in warehouses, and warehouses are actually pretty constrained," he said.

Brooks' company designed robots for practical purposes related to warehouse operations, rather than building human-looking robots. "So the form factor we use is not humanoids walking around — even though I have built and delivered more humanoids than anyone else. These look like shopping carts," he said. "It's got a handlebar, so if there's a problem with the robot, a person can grab the handlebar and do what they wish with it."

After all these years, Brooks has learned that it's about making the technology accessible and purpose-built. "I always try to make technology easy for people to understand, and therefore we can deploy it at scale, and always look at the business case; the return on investment is also very important."

Even with that, Brooks says we have to accept that there are always going to be hard-to-solve outlier cases when it comes to AI, which could take decades to solve. "Without carefully boxing in how an AI system is deployed, there is always a long tail of special cases that take decades to discover and fix. Paradoxically all those fixes are AI complete themselves."

Brooks also believes that there's a mistaken belief that technology will always grow exponentially, thanks to Moore's Law. He uses the iPod as an example. For a few iterations, it did double in storage size, but it didn't continue on that trajectory. "The models being sold in 2017 actually came with 256GB or 160GB because, as I pointed out, nobody actually needed more than that."

Brooks acknowledges that LLMs could help with domestic robots, where they could perform specific tasks, especially with an aging population and not enough people to take care of them. However, even that could come with its own set of unique challenges.

"People say, 'Oh, the large language models are gonna make robots be able to do things they couldn't do.' That's not where the problem is. The problem with being able to do stuff is about control theory and all sorts of other hardcore math optimization," he said.

Brooks believes that this could eventually lead to robots with useful language interfaces for people in care situations. "It's not useful in the warehouse to tell an individual robot to go out and get one thing for one order, but it may be useful for eldercare in homes for people to be able to say things to the robots," he said.

Image Credits: Paul Marotta / Getty Images